Radio intelligent understanding method and device based on dual-channel semantic coding deep learning
By performing high-dimensional semantic encoding on dual-channel radio signals and processing them with improved deep learning networks, the problem of low signal understanding accuracy in existing technologies has been solved, achieving more efficient signal classification and recognition.
Patent Information
- Application Number
- CN202410839461.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-26
- Publication Date
- 2025-12-26
AI Technical Summary
Existing radio signal understanding methods lack high-dimensional correlation information, resulting in low accuracy in signal understanding and an inability to efficiently and intelligently manage complex and diverse dual-channel radio signals.
A deep learning method based on dual-channel semantic coding is adopted to denoise dual-channel radio signals, perform high-dimensional semantic coding, convert the signals into two-dimensional image data using Gram angle field, and extract and classify signal features through an improved deep learning network.
It improves the accuracy of signal understanding by preserving the relationship between dual-channel signals and contextual semantic information, thus achieving more efficient signal classification and recognition.
Smart Images
Figure CN121211084A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of dual-channel radio signal, and in particular to a radio intelligent understanding method and device based on dual-channel semantic coding deep learning. BACKGROUND
[0002] With the development of communication technology, a large number of wireless communication intelligent devices are applied to our life, such as smart phones. At the same time, the demand for more efficient and flexible communication methods is also increasing, and dual-channel radio signals can provide more abundant information and more reliable transmission methods, and their applications are gradually increasing. Compared with single-channel radio signals, dual-channel radio signals are more complex and diverse.
[0003] In the face of complex and diverse dual-channel wireless signals, there are many challenges in fast and efficient intelligent management and dangerous signal discovery, and artificial intelligence technology, especially deep learning algorithm, provides an effective path to solve this problem. The intelligent understanding of radio signals is a crucial research field. The existing methods of radio signal classification, parameter estimation and technology identification mainly focus on the frequency domain, mainly using mathematical models to artificially select the frequency domain features of the signal, and then based on the selected frequency domain features, the analysis and recognition of the radio signal are realized. Intelligent understanding of radio signals. However, the existing radio signal understanding method is based on one-dimensional signal processing, lacking high-latitude correlation information, resulting in poor universality, long time consumption, low accuracy, and being unable to efficiently and intelligently manage complex and diverse dual-channel radio signals. SUMMARY
[0004] In view of the above analysis, the embodiments of the present application aim to provide a radio intelligent understanding method and device based on dual-channel semantic coding deep learning, to solve the problem of low signal understanding accuracy caused by the lack of high-latitude correlation information in the prior art.
[0005] In one aspect, the embodiments of the present application provide a radio intelligent understanding method based on dual-channel semantic coding deep learning, which comprises the following steps:
[0006] Denoising the target dual-channel radio signal;
[0007] High-dimensional semantic coding is performed on the denoised dual-channel radio signal to obtain a semantic coding signal of the dual-channel radio signal; the semantic coding signal is a two-dimensional signal including correlation;
[0008] The semantic coding signal of the dual-channel radio signal is input into a trained classification model for classification and recognition to obtain the type of the dual-channel radio signal.
[0009] Based on the further improvement of the above method, the high-dimensional semantic encoding of the denoised double-channel radio signal is performed to obtain the semantic encoding signal of the double-channel radio signal, comprising:
[0010] The first channel signal and the second channel signal of the denoised double-channel radio signal are respectively subjected to high-dimensional semantic encoding to obtain the semantic encoding signal of the first channel signal and the semantic encoding signal of the second channel signal.
[0011] The semantic encoding signal of the first channel signal and the semantic encoding signal of the second channel signal are subjected to data fusion to obtain the semantic encoding signal of the double-channel radio signal.
[0012] Based on the further improvement of the above method, the first channel signal and the second channel signal of the denoised double-channel radio signal are respectively subjected to Gram angle and field high-dimensional semantic encoding.
[0013] Based on the further improvement of the above method, the first channel signal and the second channel signal of the denoised double-channel radio signal are respectively subjected to Gram angle and field high-dimensional semantic encoding, comprising:
[0014] The first channel signal and the second channel signal are respectively taken as target signals.
[0015] The signal value of the target signal is scaled to the interval [0, 1] to obtain a normalized signal.
[0016] The normalized signal is converted into polar coordinate representation to obtain the polar coordinates corresponding to each signal value.
[0017] The Gram angle field matrix is calculated according to the angle value of the polar coordinates of each signal value, and the Gram angle field matrix is taken as the semantic encoding of the target signal.
[0018] Based on the further improvement of the above method, the normalized signal is converted into polar coordinate representation according to the following formula:
[0019] φ i =arccos(x i "),-1≤x i "≤1,x i "∈S';
[0020] S'={x1",x2",...x n "};
[0021]
[0022] In the formula, S' is the time sequence of the normalized signal, i is the time sequence number, i = 1...n, ti is the time stamp, x i is t i is the signal value at time point, N is the sum of time stamps, φ i is t i is the angle value of polar coordinates at time point, r i is t i is the radius of polar coordinates at time point.
[0023] Based on the further improvement of the above method, the formula for calculating the Gram angular field matrix according to the angle value of the polar coordinates of each signal value is:
[0024] GASF ij = cos(φ i + φ j );
[0025] In the formula, GASF ij is the element in the i-th row and j-th column of the Gram angular field matrix, φ i is the angle value of the polar coordinates at time point t i is the angle value of the polar coordinates at time point t j is the angle value of the polar coordinates at time point t j is the angle value of the polar coordinates at time point t, i, j = 1…n.
[0026] Based on the further improvement of the above method, the semantic encoding signals of the first channel signal and the semantic encoding signals of the second channel signal are data fused, and the formula is:
[0027]
[0028] In the formula, GASF S,ij is the signal value in the i-th row and j-th column of the semantic encoding signal of the dual-channel radio signal, GASF I,ij is the signal value in the i-th row and j-th column of the first channel signal, GASF Q,ij is the signal value in the i-th row and j-th column of the second channel signal.
[0029] Based on the further improvement of the above method, the denoising processing of the target dual-channel radio signal includes:
[0030] The dual-channel radio signal is segmented and summed for smoothing by using a non-overlapping sliding window of a preset size to remove signal noise.
[0031] Based on the further improvement of the above method, the classification model includes:
[0032] A first convolutional layer performs a two-dimensional convolution operation on the input signal data;
[0033] A first pooling layer performs an average pooling operation on the output data of the first convolutional layer;
[0034] a first normalization layer, configured to perform a normalization operation on output data of the first pooling layer;
[0035] a first activation layer, configured to perform an activation operation on output data of the first normalization layer by using a ReLU activation function;
[0036] a second convolution layer, configured to perform a two-dimensional convolution operation on output data of the first activation layer;
[0037] a second normalization layer, configured to perform a normalization operation on output data of the second convolution layer;
[0038] a second activation layer, configured to perform an activation operation on output data of the second normalization layer by using a ReLU activation function;
[0039] a second pooling layer, configured to perform a maximum value pooling operation on output data of the second activation layer;
[0040] a flattening layer, configured to perform a flattening operation on output data of the second pooling layer to obtain one-dimensional signal data;
[0041] a full connection layer, configured to perform a full connection layer operation on the one-dimensional signal data output by the flattening layer to obtain a type of the dual-channel radio signal.
[0042] In another aspect, an embodiment of the present application provides a radio intelligent understanding device based on dual-channel semantic coding deep learning, which comprises:
[0043] a signal acquisition module, configured to acquire a dual-channel radio signal;
[0044] a denoising module, configured to perform denoising processing on the dual-channel radio signal;
[0045] a semantic coding module, configured to perform high-dimensional semantic coding on the dual-channel radio signal after denoising processing to obtain a semantic coding signal of the dual-channel radio signal;
[0046] a classification model, configured to perform classification identification on the semantic coding signal of the dual-channel radio signal to obtain a type of the dual-channel radio signal.
[0047] Compared with the prior art, the present application can achieve at least one of the following beneficial effects:
[0048] 1. In the present application, the dual-channel radio signal is subjected to high-dimensional semantic coding, high-dimensional signal semantic association is established, the one-dimensional dual-channel radio signal is converted into a high-dimensional semantic coding signal, and then the automatic selection and extraction of signal features are performed through the classification model, so that the understanding of the radio signal is realized by using high-dimensional information, and the accuracy of signal understanding is improved.
[0049] 2、The application discloses a radio intelligent understanding method based on double-channel semantic coding deep learning.
[0050] 3、The application discloses a radio intelligent understanding method based on double-channel semantic coding deep learning.
[0051] The technical solutions in the application can be combined with each other to realize more preferred combination solutions. BRIEF DESCRIPTION OF DRAWINGS
[0052] The drawings are only used for the purpose of illustrating specific embodiments and are not considered as limiting the application, and the same reference signs represent the same components throughout the drawings.
[0053] Figure 1 The application discloses a radio intelligent understanding method based on double-channel semantic coding deep learning. DETAILED DESCRIPTION
[0054] The preferred embodiments of the application are described below in conjunction with the drawings, which form a part of this application and are used to illustrate the principles of the application, and are not considered as limiting the scope of the application.
[0055] One specific embodiment of the application discloses a radio intelligent understanding method based on double-channel semantic coding deep learning, as shown in Figure 1 The method comprises the following steps:
[0056] Step 1, denoising the target double-channel radio signal;
[0057] Step 2, high-dimensional semantic coding of the denoised double-channel radio signal to obtain the semantic coding signal of the double-channel radio signal; the semantic coding signal is a two-dimensional signal including an association relationship;
[0058] Step 3, input the semantic coding signal of the dual-channel radio signal into the trained classification model for classification and recognition to obtain the type of the dual-channel radio signal.
[0059] Compared with the prior art, in the embodiment of the present application, the dual-channel radio signal is high-dimensionally semantically coded, high-dimension signal semantic correlation is established, the one-dimensional dual-channel radio signal is converted into a high-dimension semantic coding signal, and then the automatic selection and extraction of signal features are performed through a classification model, so that the understanding of the radio signal is realized by using high-dimension information, and the accuracy of signal understanding is improved.
[0060] It should be noted that the dual-channel radio signal refers to a signal containing two independent channels, which are defined as a first channel signal and a second channel signal. Meanwhile, the dual-channel radio signal in the embodiment of the present application is a time sequence signal.
[0061] Step 1, denoising processing is performed on the target dual-channel radio signal.
[0062] Specifically, the dual-channel radio signal is segmented and summed for smoothing by using a non-overlapping sliding window of a preset size to remove signal noise.
[0063] In the embodiment of the present application, the signal-to-noise ratio of the signal can be effectively improved by segmenting and summing for smoothing, the signal quality is improved, and the noise is removed. The sliding window is used to extract a signal segment. Through the sliding window, the signal is divided into multiple small segments. By using a non-overlapping sliding window, it can be ensured that each part of the signal is independently processed, and the original structure of the signal can be preserved. When segmenting and summing for smoothing, the values of the signal in each window are added up, which is conducive to emphasizing the strong components in the signal and reducing random noise.
[0064] It should be noted that the signals of the two channels of the dual-channel radio signal, i.e. the first channel signal and the second channel signal, are respectively segmented and summed for smoothing.
[0065] The original signal to be denoised is defined as S0={x1,x2,...x m}. The formula for segmenting and summing for smoothing the dual-channel radio signal by using a non-overlapping sliding window of a preset size is:
[0066]
[0067] In the formula, i is a time sequence number, x′ i is the smoothed signal value after segmenting and summing in the i-th sliding window, x' i ∈S; the signal values in the i-th sliding window start from x i and end at x i+span-1 ; and span is the window size.
[0068] In implementation, span = 2. After the original signal is segmented and summed and smoothed, the signal after denoising is converted into S = {x' 1, x' 2,..., x' n}. n}.
[0069] Step 2, high-dimensional semantic coding is performed on the double-channel radio signal after denoising, and a semantic coding signal of the double-channel radio signal is obtained.
[0070] Specifically, the high-dimensional semantic coding is performed on the double-channel radio signal after denoising, and the semantic coding signal of the double-channel radio signal is obtained, including:
[0071] The first channel signal and the second channel signal of the double-channel radio signal after denoising are respectively subjected to high-dimensional semantic coding, and the semantic coding signal of the first channel signal and the semantic coding signal of the second channel signal are obtained.
[0072] The semantic coding signal of the first channel signal and the semantic coding signal of the second channel signal are subjected to data fusion, and the semantic coding signal of the double-channel radio signal is obtained.
[0073] Compared with the prior art, in the embodiment of the application, when high-dimensional semantic coding is performed on the double-channel radio signal, the relationship between the double-channel signals is considered, and fusion calculation after conversion of the double-channel signals is proposed. After high-dimensional semantic coding is performed on the first channel signal and the second channel signal respectively, the semantic coding signal of the first channel signal and the semantic coding signal of the second channel signal are fused, so that more useful context semantic information in the original signal is retained, and the accuracy of signal understanding is improved.
[0074] In one specific embodiment, Gram angle field is used to perform high-dimensional semantic coding on the first channel signal and the second channel signal of the double-channel radio signal after denoising.
[0075] Gram angle field (GAF) is a method of converting one-dimensional time series data into two-dimensional image data. This method converts time series in Cartesian coordinate system into angle and radius in polar coordinate system through polar coordinate coding, and then generates a Gram angle field matrix, realizing one-dimensional to two-dimensional conversion.
[0076] The Gram angle field can better retain the global characteristics of time series data by calculating the dot product and cosine similarity, and is suitable for capturing nonlinear relationships in time series. Unlike time-frequency diagrams, the Gram angle field does not need to perform frequency domain transformation, and is therefore not affected by the selection of frequency domain resolution, and can more robustly represent signals in different frequency ranges. In addition, the Gram angle field can convert time series data into image data, not only retaining the complete information of the signal, but also maintaining the signal's dependence on time, facilitating image classification and recognition using deep learning models such as neural networks.
[0077] Specifically, the Gram angle field is used to perform high-dimensional semantic encoding on the first channel signal and the second channel signal of the denoised dual-channel radio signal, respectively, including:
[0078] Respectively, the first channel signal and the second channel signal are taken as target signals;
[0079] The signal values of the target signals are scaled to the interval [0, 1] to obtain normalized signals;
[0080] The normalized signals are converted into polar coordinate representation to obtain polar coordinates corresponding to each signal value;
[0081] The Gram angle sum field matrix is calculated according to the angle value of the polar coordinates of each signal value, and the Gram angle sum field matrix is taken as the semantic encoding signal value of the target signal.
[0082] It should be noted that the Gram angle field includes the Gram angle sum field and the Gram angle difference field, and in the embodiment of the present application, the Gram angle sum field is used for high-dimensional semantic encoding.
[0083] Specifically, the normalized signal is converted into polar coordinate representation according to the following formula:
[0084] φ i =arccos(x i "),-1≤x i "≤1,x i "∈S';
[0085] S'={x1",x2",...x n "};
[0086]
[0087] In the formula, S' is the time series of the normalized signal, i is the time serial number, i = 1...n, t i is the time stamp, x i " is the signal value at the t i time point, N is the sum of the time stamps, φ i is the ti the angle value of the polar coordinates of the time point, r i is t i the radius of the polar coordinates of the time point.
[0088] In this embodiment, the cosine value of the normalized signal is calculated as the angle of the polar coordinates, and the radius of the polar coordinates is calculated according to the timestamp of the normalized signal.
[0089] Specifically, the formula for calculating the Gram angle sum field matrix according to the angle value of the polar coordinates of each signal value is:
[0090] In the formula, GASF ij is the element in the i-th row and the j-th column of the Gram angle sum field matrix, φ i is t i the angle value of the polar coordinates of the time point, φ j is t j the angle value of the polar coordinates of the time point, i, j = 1…n.
[0091] Specifically, the semantic encoding signals of the first channel signal and the semantic encoding signals of the second channel signal are data fused, and the formula is:
[0092]
[0093] In the formula, GASF S,ij is the signal value in the i-th row and the j-th column of the semantic encoding signal of the dual-channel radio signal, GASF I,ij is the signal value in the i-th row and the j-th column of the first channel signal, GASF Q,ij is the signal value in the i-th row and the j-th column of the second channel signal.
[0094] It should be noted that the semantic encoding signal of the embodiment of the application is a two-dimensional signal.
[0095] Step 3, input the semantic encoding signal of the dual-channel radio signal into the trained classification model for classification and recognition to obtain the type of the dual-channel radio signal.
[0096] Specifically, the classification model comprises:
[0097] a first convolutional layer for performing a two-dimensional convolution operation on the input signal data;
[0098] a first pooling layer for performing an average pooling operation on the output data of the first convolutional layer;
[0099] a first normalization layer for performing a normalization operation on the output data of the first pooling layer;
[0100] The first activation layer adopts a ReLU activation function to activate output data of the first normalization layer;
[0101] The second convolution layer performs a two-dimensional convolution operation on the output data of the first activation layer.
[0102] The second normalization layer performs a normalization operation on the output data of the second convolution layer.
[0103] The second activation layer adopts a ReLU activation function to activate output data of the second normalization layer.
[0104] The second pooling layer performs a maximum value pooling operation on the output data of the second activation layer.
[0105] The flattening layer performs a flattening operation on the output data of the second pooling layer to obtain one-dimensional signal data.
[0106] The full connection layer performs a full connection layer operation on the one-dimensional signal data output by the flattening layer to obtain the type of the dual-channel radio signal.
[0107] Compared with the prior art, in the embodiment of the present application, a classification model is constructed based on a deep learning network, and the model structure is improved, the classification model can intelligently and automatically extract rich features of high-dimensional signals, and is beneficial to improve the accuracy of signal understanding.
[0108] Specifically, in the embodiment of the present application, two groups of convolution layers, pooling layers, normalization layers and activation layers are used to extract features from input data twice to improve classification accuracy, wherein in the first feature extraction, the first pooling layer adopts average pooling to retain global information and avoid losing semantic correlation information, and in the second feature extraction, the second pooling layer adopts maximum pooling to reduce the spatial dimension of data, retain local features and highlight the most important features, which is beneficial to the classification model to capture semantic correlation information in signal data, thereby improving the accuracy of signal understanding.
[0109] In specific implementation, the first convolution layer performs a two-dimensional convolution operation on input signal data, the input of the first convolution layer is 1 channel, the output is 8 channels, the size of the convolution kernel is 3*3, and the step is 1.
[0110] The first pooling layer performs an average pooling operation of 2*2 average value on the output data of the 8 channels output by the first convolution layer.
[0111] The first normalization layer performs a normalization operation in the range of [0-1] on the output data of the 8 channels output by the first pooling layer.
[0112] The first activation layer respectively activates the output data of the 8 channels output by the first normalization layer using a ReLU activation function.
[0113] The second convolution layer performs a two-dimensional convolution operation on the output data of the 8 channels output by the first activation layer again, the input of the second convolution layer is 8 channels, the output is 16 channels, the convolution kernel size is 3*3, and the step is 1. When performing the two-dimensional convolution operation, a circle of preset information values is filled along the image edge, so that the convolution kernel performs convolution operation on the data of the image edge.
[0114] The second normalization layer respectively performs a normalization operation in the range of [0-1] on the output data of the 16 channels output by the second convolution layer.
[0115] The second activation layer respectively activates the output data of the 16 channels output by the second normalization layer using a ReLU activation function.
[0116] The second pooling layer performs a maximum value pooling operation of 2*2 maximum value on the output data of the 16 channels output by the second activation layer.
[0117] The flattening layer performs a tiling operation on the output data of the 16 channels output by the second pooling layer to form a one-dimensional signal data with a length of 4096.
[0118] The fully connected layer performs a full connection operation on the one-dimensional signal data with a length of 4096 output by the flattening layer, and outputs a signal category.
[0119] Preferably, the semantic encoding signal of the dual-channel radio signal is converted into an image and input into the trained classification model. Wherein the semantic encoding signal is a two-dimensional signal, and when converted into an image, each semantic encoding signal value corresponds to the gray value or RGB value of the corresponding position in the image.
[0120] The classification model used in the embodiment of the application is a trained model. Specifically, the classification model is trained by the following method:
[0121] Step A1, collecting dual-channel radio signal data of a target application scenario.
[0122] Step A2, dividing the dual-channel radio signal data into a training set and a test set according to a preset ratio; for example, dividing the training set and the test set according to an 8:2 ratio.
[0123] Step A3, respectively performing denoising processing on the dual-channel radio signal data in the training set and the test set.
[0124] Step A4, high-dimensional semantic coding is performed on the denoised double-channel radio signal, and semantic coding signals of the double-channel radio signal of the training set and the test set are obtained respectively.
[0125] Step A5, the classification model is trained by using the semantic coding signals of the training set, and the classification accuracy of the trained classification model is verified by using the semantic coding signals of the test set.
[0126] In another aspect, the embodiment of the present application also provides a radio intelligent understanding device based on double-channel semantic coding deep learning. The device comprises:
[0127] A denoising module is configured to perform denoising processing on a target double-channel radio signal.
[0128] A semantic coding module is configured to perform high-dimensional semantic coding on the denoised double-channel radio signal, and obtain a semantic coding signal of the double-channel radio signal.
[0129] A classification model is configured to perform classification and identification on the semantic coding signal of the double-channel radio signal, and obtain a type of the double-channel radio signal.
[0130] The radio intelligent understanding device of the embodiment of the present application is used to implement the radio intelligent understanding method of the aforementioned embodiment. The radio intelligent understanding method and device of the embodiment of the present application can achieve the same technical effects, which will not be described here.
[0131] Those skilled in the art can understand that all or part of the processes of the above-mentioned embodiments can be completed by a computer program instructing related hardware, and the program can be stored in a computer readable storage medium. The computer readable storage medium includes a magnetic disk, an optical disk, a read-only memory, a random access memory, etc.
[0132] The above description is only a preferred embodiment of the present application, but the protection scope of the present application is not limited to this. Any changes or replacements within the technical scope disclosed by the present application can be easily thought by those skilled in the art, which should be covered within the protection scope of the present application.
Claims
1. A radio intelligent understanding method based on dual-channel semantic coding deep learning, characterized in that, The method includes the following steps: Denoising the target's dual-channel radio signals; The denoised dual-channel radio signal is subjected to high-dimensional semantic coding to obtain the semantically coded signal of the dual-channel radio signal; the semantically coded signal is a two-dimensional signal including correlation relationships; The semantically encoded signal of the dual-channel radio signal is input into a trained classification model for classification and identification to obtain the type of the dual-channel radio signal.
2. The method according to claim 1, characterized in that, The step of performing high-dimensional semantic coding on the denoised dual-channel radio signal to obtain the semantically coded signal of the dual-channel radio signal includes: High-dimensional semantic coding is performed on the first channel signal and the second channel signal of the denoised dual-channel radio signal respectively to obtain the semantic coding signal of the first channel signal and the semantic coding signal of the second channel signal; The semantic encoded signals of the first channel signal and the second channel signal are fused to obtain the semantic encoded signal of the dual-channel radio signal.
3. The method according to claim 2, characterized in that, The first and second channel signals of the denoised dual-channel radio signal are respectively subjected to high-dimensional semantic encoding using Gram angle field.
4. The method according to claim 3, characterized in that, The step of using Gram angle field to perform high-dimensional semantic encoding on the first and second channel signals of the denoised dual-channel radio signal includes: The first channel signal and the second channel signal are respectively used as target signals; The signal value of the target signal is scaled to the interval [0,1] to obtain a normalized signal; The normalized signal is converted into polar coordinates to obtain the polar coordinates corresponding to each signal value. The Gram angle and field matrix are calculated based on the polar coordinate angle values of each signal value, and the Gram angle and field matrix are used as the semantic encoding signal of the target signal.
5. The method according to claim 4, characterized in that, The normalized signal is converted to polar coordinates using the following formula: φ i =arccos(x″ i ),-1≤x″ i ≤1,x″ i ∈S'; S'={x″1,x″2,...x″ n }; In the formula, S' is the normalized time series of the signal, i is the time index, i = 1…n, t i For timestamps, x″ i For t i The signal value at a given time point, where N is the sum of timestamps, and φ i For t i The angle value of the polar coordinates at a given time point, r i For t i The radius of the polar coordinates at a given time point.
6. The method according to claim 5, characterized in that, The formula for calculating the Gram angle and field matrix based on the polar coordinate angle value of each signal value is as follows: GASF ij =cos(φ i +φ j ); In the formula, GASF ij Let φ be the element in the i-th row and j-th column of the Gram angle and field matrix. i For t i The angle value of the polar coordinates at a given time point, φ j For t j The polar coordinate angle values at time points, i,j=1…n.
7. The method according to claim 6, characterized in that, The semantically encoded signals of the first channel signal and the second channel signal are fused using the following formula: In the formula, GASF S,ij GASF represents the signal value in the i-th row and j-th column of the semantically encoded signal of a dual-channel radio signal. I,ij Let GASF be the signal value in the i-th row and j-th column of the first channel signal. Q,ij This represents the signal value in the i-th row and j-th column of the second channel signal.
8. The method according to any one of claims 1-7, characterized in that, The noise reduction processing of the target dual-channel radio signal includes: The dual-channel radio signals are segmented and smoothed using a non-overlapping sliding window of a preset size to remove signal noise.
9. The method according to any one of claims 1-7, characterized in that, The classification model includes: The first convolutional layer performs a two-dimensional convolution operation on the input signal data. The first pooling layer performs average pooling on the output data of the first convolutional layer. The first normalization layer performs a normalization operation on the output data of the first pooling layer; The first activation layer uses the ReLU activation function to activate the output data of the first normalization layer; The second convolutional layer performs a two-dimensional convolution operation on the output of the first activation layer. The second normalization layer performs a normalization operation on the output data of the second convolutional layer; The second activation layer uses the ReLU activation function to activate the output data of the second normalization layer; The second pooling layer performs max pooling on the output data of the second activation layer; A flattening layer is used to flatten the output data of the second pooling layer to obtain one-dimensional signal data. A fully connected layer performs fully connected layer operations on the one-dimensional signal data output by the flattened layer to obtain the type of the dual-channel radio signal.
10. A radio intelligent understanding device based on dual-channel semantic coding deep learning, characterized in that, The device includes: The signal acquisition module acquires dual-channel radio signals; A noise reduction module performs noise reduction processing on the dual-channel radio signals; The semantic coding module performs high-dimensional semantic coding on the denoised dual-channel radio signal to obtain the semantically coded signal of the dual-channel radio signal; A classification model is used to classify and identify the semantically encoded signals of the dual-channel radio signals to obtain the type of the dual-channel radio signals.